Remix.run Logo
TPU Deep Dive(henryhmko.github.io)
235 points by transpute 11 hours ago | 25 comments
lanthissa 2 hours ago | parent | next [-]

can someone help me understand how the following can be true:

1. TPU's are a serious competitor to nvidia chips.

2. Chip makers with the best chips are valued at 1-3.5T.

3. Google's market cap is 2T.

4. It is correct for google to not sell TPU's.

i have heard the whole, its better to rent them thing, but if they're actually good selling them is almost as good a business as every other part of the company.

Velorivox 22 minutes ago | parent | next [-]

Wall street undervalued Google even on day one (IPO). Bezos has said that some of the times the stock had been doing the worst were when the company was doing great.

So, to help you understand how they can be true: market cap is governed by something other than what a business is worth.

radialstub an hour ago | parent | prev | next [-]

I believe Broadcom is also very involved in the making of the TPU's and networking infrastructure and they are valued at 1.2T currently. Maybe consider the combined value of Broadcom and Google.

mft_ an hour ago | parent | prev | next [-]

If they think they’ve got a competitive advantage vs. GPUs which benefits one of their core products, it would make sense to retain that competitive advantage for the long term, no?

smokel 2 hours ago | parent | prev | next [-]

Selling them and supporting that in the field requires quite some infrastructure you'd have to build. Why go through all that trouble if you already make higher margins renting them out?

Also, if they are so good, it's best to not level the playing field by sharing that with your competitors.

Also "chip makers with the best chips" == Nvidia, there aren't many others. And Alphabet does more than just produce TPUs.

rwmj an hour ago | parent | prev | next [-]

Aren't Google's TPUs a bit like a research project with practical applications as a nice side effect?

dismalaf an hour ago | parent | prev [-]

Nvidia is selling a ton of chips on hype.

Google is saving a ton of money by making TPUs, which will pay off in the future when AI is better monetized, but so far no one is directly making a massive profit from foundation models. It's a long term play.

Also, I'd argue Nvidia is massively overvalued.

RossBencina 7 hours ago | parent | prev | next [-]

Can you suggest a good reference for understanding which algorithms map well onto the regular grid systolic arrays used by TPUs? The fine article says dese matmul and convolution are good, but is there anything else? Eigendecomposition? SVD? matrix exponential? Solving Ax = b or AX = B? Cholesky?

cdavid 5 hours ago | parent | next [-]

SVD/eigendecomposition will often boil down to making many matmul (e.g. when using Krylov-based methods, e.g. Arnoldi, Krylov-schur, etc.), so I would expect TPU to work well there. GMRES, one method to solve Ax = b is also based on Arnoldi decomp.

WithinReason 7 hours ago | parent | prev | next [-]

Anything that you can express as 128x128 (but ideally much larger) dense matrix multiplication and nothing else

musebox35 7 hours ago | parent | prev [-]

I think https://jax-ml.github.io/scaling-book/ is one of the best references to go through. It details how single device and distributed computations map to TPU hardware features. The emphasis is on mapping the transformer computations, both forwards and backwards, so requires some familiarity with how transformer networks are structured.

cdg007 20 minutes ago | parent | prev | next [-]

What will competitors say?

serf 7 hours ago | parent | prev | next [-]

does that cooling channel have a NEMA stepper on it as a pump or metering valve?[0]

If so, wild. That seems like overkill.

[0]: https://henryhmko.github.io/posts/tpu/images/tpu_tray.png

fellowmartian 5 hours ago | parent [-]

definitely closed-loop, might even be a servo

sgt101 5 hours ago | parent | prev | next [-]

ELI5: how (specifically) do GPU and TPU optimisations effect determinism in LLMs? Or is this all a myth?

barrkel 4 hours ago | parent | next [-]

LLMs are generally deterministic. The token sampling step is usually randomized to some degree because it gets better results (creativity) and helps avoid loops, but you can turn that off (temp zero for simple samplers).

sgeisenh 34 minutes ago | parent | next [-]

This is an oversimplification. When distributed, the nondeterministic order of additions during reductions can produce nondeterministic results due to floating point error.

It’s nitpicking for sure, but it causes real challenges for reproducibility, especially during model training.

perching_aix 2 hours ago | parent | prev [-]

+ can also just pin the seed instead, right?

jpgvm 3 hours ago | parent | prev [-]

They don't affect determinism of the results but different architectures have different determinism guarantees with respect to performance, as a result of scheduling and other things.

TPUs share a similar lineage to the Groq TPU accelerators (disclaimer: I work at Groq) which are actually fully deterministic which means not only do you get deterministic output, you get it in a deterministic number of cycles.

There is a trade off though, making the hardware deterministic means you give up HW level scheduling and other sources of non-determinism. This makes the architecture highly dependent on a "sufficiently smart compiler". TPUs and processors like them are generally considered VLIW and are all similarly dependent on the compiler doing all the smart scheduling decisions upfront to ensure good compute/IO overlap and eliminating pipeline bubbles etc.

GPUs on the other hand have very sophisticated scheduling systems on the chips themselves along with stuff like kernel swapping etc that make them much more flexible, less dependent on the compiler and generally easier to reach a fairly high utilisation of the processor without too much work.

TLDR: TPUs MAY have deterministic cycle guarantees. GPUs (of the current generation/architectures) cannot because they use non-deterministic scheduling and memory access patterns. Both still produce deterministic output for deterministic programs.

almostgotcaught 9 hours ago | parent | prev | next [-]

> In essence, caches allow hardware to be flexible and adapt to a wide range of applications. This is a large reason why GPUs are very flexible hardware (note: compared to TPUs).

this is correct but mis-stated - it's not the caches themselves that cost energy but MMUs that automatically load/fetch/store to cache on "page faults". TPUs don't have MMUs and furthermore are a push architecture (as opposed to pull).

frays 6 hours ago | parent | prev | next [-]

How can someone have this level of knowledge about TPUs without working at Google?

ipsum2 5 hours ago | parent | next [-]

Everything thats in the blog post is basically well known already. Google publishes papers and gives talks about their TPUs. Many details are lacking though, and require some assumptions/best guesses. Jax and XLA are (partially) open source and give clues about how TPUs work under the hood as well.

https://arxiv.org/abs/2304.01433

https://jax-ml.github.io/scaling-book/

musebox35 5 hours ago | parent | prev [-]

From the acknowledgment at the end, I guess the author has access to TPUs through https://sites.research.google/trc/about/

This is not the only way though. TPUs are available to companies operating on GCP as an alternative to GPUs with a different price/performance point. That is another way to get hands-on experience with TPUs.

erwincoumans 4 hours ago | parent [-]

A quick free way to access TPUs is through https://colab.research.google.com, Runtime / Change Runtime Type / v2-8 TPU

ariwilson 5 hours ago | parent | prev [-]

Cool article!